Artificial intelligence is already writing your emails. It’s drafting business memos, summarizing meetings, assisting in legal briefs, composing customer service scripts, and even ghostwriting books. But amid all this futuristic promise, one peculiar fact stands out: AI still talks like it’s stuck in a dusty library full of old novels, wartime telegrams, and undergraduate term papers from 1997.

Spend a few minutes with a popular AI tool like ChatGPT, Claude, or Gemini, and you’ll soon encounter phrases that feel oddly… anachronistic. Words like “testament,” “peril,” “delve,” and “henceforth” are sprinkled into otherwise modern contexts. The AI might speak of someone “clutching their pearls” in surprise, or “padding softly” through a hallway, as if we’re in a noir detective novel rather than a Slack thread.

This isn’t just a cute bug. It’s a symptom of something deeper—something structural—about how AI sees the world through language. And it reveals a surprising truth about how the past continues to haunt our most advanced technologies.

Statistical Brains, Not Storytellers

To understand why AI talks the way it does, you first need to understand what it actually is. Large language models (LLMs) don’t reason. They don’t plan. They don’t think in any conventional sense. They’re statistical pattern matchers: predictive engines trained to guess the next most likely word based on billions of text samples.

And what kind of texts have they read? A whole lot of old stuff.

Much of the material used to train these models includes:

• Public domain books: Think 19th- and early 20th-century literature, legal texts, religious writings, and philosophy—all free, all plentiful, all polished in a way that screams “trustworthy” to an algorithm.
• Government reports and academic articles: Also in the public domain. Also formal. Also ancient by internet standards.
• Wikipedia, blogs, forums, and scraped websites: Some contemporary, some wildly outdated, often uneven in tone and relevance.

So when you ask an AI to draft an article or tell a story, it pulls from a stew of language that’s heavily seasoned with bygone phrasing. In short: AI doesn’t know it sounds old-fashioned, because it doesn’t know anything at all. But it has seen phrases like “grave peril” or “a testament to their courage” used millions of times in its training data. So it repeats them, assuming they’re what we want.

A Formality Crisis in the Age of Casual Speech

What’s so bad about sounding a little old-school? Isn’t it better than sounding sloppy?

Maybe. But in our modern world—especially online—authenticity often means casualness. Our writing is full of contractions, emoji, memes, ironic asides, and tone that shifts on a dime. Contemporary communication values voice and spontaneity over formality and polish.

AI doesn’t quite get that. It often over-corrects for professionalism, defaulting to the tone of a stiff academic, an over-eager intern, or a particularly verbose customer service rep.

Consider this sentence:

“In light of recent developments, it is imperative that we remain vigilant.”

It’s perfectly correct. It’s also something very few people would actually say in a normal conversation. But AI loves these kinds of constructions because they’re predictable, formal, and appear frequently across corporate and governmental documents.

The result? Text that’s technically flawless but emotionally hollow. It’s a voice without a soul—something that sounds like writing but doesn’t feel like a human wrote it.

This Isn’t a Bug—It’s a Mirror

We often talk about AI’s flaws as if they’re separate from us. But the formality, the stiffness, the fondness for archaic turns of phrase—these aren’t purely machine errors. They’re reflections of what we, collectively, have written, digitized, and made available. The AI isn’t being weird. It’s just being us, but averaged out, decontextualized, and boiled down to patterns.

And that’s worth thinking about.

Because in many ways, the formal, generic, emotionally distant tone of AI writing mirrors the bureaucratic, impersonal, and often inaccessible tone that’s dominated institutional communication for decades. When you ask an AI for help, you’re getting the accumulated language of textbooks, manuals, white papers, and Wikipedia entries. You’re not getting jazz. You’re getting the elevator music of prose.

Can AI Ever Really Talk Like Us?

Developers are working on it. Newer models are trained on fresher, more conversational data. Some systems let you specify tone (“make it sound like a TikTok influencer” or “write like a sarcastic roommate”). That helps.

But without a true understanding of culture, humor, subtext, or audience, AI remains a mimic, not a messenger. It can reflect a tone it’s been shown, but it can’t invent one. It can generate jokes, but it doesn’t know why they’re funny. It can write stories, but it doesn’t feel suspense. It can sound casual, but only if you tell it to.

Even then, it may still say something like:

“She paused, clutching her pearls in disbelief…”

And you’ll sigh, because you’re not writing a 1938 radio play.

Final Thoughts: What We Risk When We Accept the Voice of the Machine

Here’s the real danger: As we increasingly rely on AI to generate content, there’s a risk we’ll unconsciously adopt its tone as the new norm. Already, AI-generated emails, help docs, blog posts, and press releases are seeping into the world—and they’re making our collective voice flatter, duller, and more robotic.

If we’re not careful, we could end up sounding more like the machine than the machine sounds like us.

The antidote isn’t to reject AI. It’s to edit ruthlessly, to train intentionally, and to keep human voice at the center of communication. We can use AI to jump-start our work—but we shouldn’t let it replace the messiness, creativity, and cultural specificity that make writing feel real.

Because in the end, language isn’t just about conveying information. It’s about connection. And if we let the machine do all the talking, we may find that the voices we lose… were our own.